Translation carried out by a machine, as you may have realized by now, is very much a problem of a probabilistic nature. A good result depends exclusively on probabilities.
That’s why a recent trend in translation techniques is to design translation programs based on statistical comparison of the original text against pre-translated texts… huge amounts of text, perhaps millions of pages.
Within the Internet, you may find some of these statistical translators offered as a free service. The advantage is that sometimes you obtain perfect, human-like translations. The disadvantage is that you also have to “train” your system with millions of pages of documents pre-translated by humans. Another disadvantage is that if you change the wording of your original text, even in the slightest way, you may obtain surprisingly different results. That’s because the modified sentence is not—perhaps—part of the “standard” pre-translated database (called corpus).
Here is the same example we used in our last tip. Try it in one of the statistical translators available on the Internet and you will see that you get a perfect translation.
However, change the tense of the verb and see what happens!
Or in the future tense:
Notice that there is no consistency between one result and the other. The other translations do not even render the correct tense.
In statistical translation, we not only have the problem of idioms and colloquial expressions to deal with, but also one has to be careful to feed in just those sentences which have been previously translated in the corpus (database). A slight change will produce huge distortions. The problems are:
How do you know which sentences are already included as pre-translated?
If you obtain a bad translation, how do you fix it?
How do you know if you are getting a good translation in the first place, in case you are not fluent in the other language?
Our Interactive Professional Text Translator takes care of all these setbacks. Click Here to download a free trial of our Text Translator.